------- Comment #6 from aoliva at gcc dot gnu dot org 2010-06-02 06:28 ------- Andrew, the debug stmts are indeed originally from the loop header's PHI nodes. Before ch we have:
i_1 = <init>; # i => i_1 goto <header>; <body>: i_3 = <whatever>(i_2); # i => i_3 <header>: i_2 = PHI <i_1, i_3>; # i => i_2 if (<cond>) goto <body>; else goto <exit>; copy_loop_header turns this into: i_1 = <init>; # i => i_1 <copy>: i_4 = PHI <i_1>; # i => i_4 if (<cond>) goto <body>; else goto <exit>; <body>: i_3 = <whatever>(i_2); # i => i_3 <header>: i_2 = PHI <i_3>; # i => i_2 if (<cond>) goto <body>; else goto <exit>; and that's correct. Then cfgcleanup realizes the headers can be merged with their preds, and that kills the PHI nodes, but there's no reason why this should kill the debug stmts. Of course we could arrange for some cleanup passes to detect and remove debug stmts in case they become redundant, but I don't see that we have a bug here. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43222
